You can’t burn the house down because of one bedbug: a qualitative study of changing gender norms in the prevention of violence against women and girls in an urban informal settlement in India

Background: The contribution of structural inequalities and societal legitimisation to violence against women, which 30% of women in India survive each year, is widely accepted. There is a consensus that interventions should aim to change gender norms, particularly through community mobilisation. How this should be done is less clear. Methods: We did a qualitative study in a large informal settlement in Mumbai, an environment that characterises 41% of households. After reviewing the anonymised records of consultations with 1653 survivors of violence, we conducted 5 focus group discussions and 13 individual interviews with 71 women and men representing a range of age groups and communities. We based the interviews on fictitious biographical vignettes to elicit responses and develop an understanding of social norms. We wondered whether, in trying to change norms, we might exploit the disjunction between descriptive norms (beliefs about what others actually do) and injunctive norms (beliefs about what others think one ought to do), focusing program activities on evidence that descriptive norms are changing. Results: We found that descriptive and injunctive norms were relatively similar with regard to femininity, masculinity, the need for marriage and childbearing, resistance to separation and divorce, and disapproval of friendships between women and men. Some constraints on women’s dress and mobility were relaxing, but there were more substantial differences between descriptive and injunctive norms around women’s education, control of income and finances, and premarital sexual relationships. Conclusions: Programmatically, we hope to exploit these areas of mismatch in the context of injunctive norms generally inimical to violence against women. We propose that an under-appreciated strategy is expansion of the reference group: induction of relatively isolated women and men into broader social groups whose descriptive and injunctive norms do not tolerate violence


Introduction
Across the world, women continue to suffer physical, emotional, sexual, and economic violence 1 . Preventing such violence has been a World Health Organization priority since 2013 2 , and is a target for the fifth Sustainable Development Goal. The contribution to endemic violence of structural inequalities and implicit and explicit legitimisation is now widely accepted 3 . Contemporary prevention programs attempt to address determinants such as patriarchal arrangements, hegemonic masculinity, and inequitable gender roles. A common way to understand the determinants of violence against women and girls is to frame them in a socioecological model that locates individual personal histories within families, located in turn within communities, and in turn within societies 4,5 . There is broad agreement that interventions should operate at multiple levels, from individual to societal. Interventional discourse has also moved along this path, from a concentration on the needs of survivors of violence to an acknowledgment that intervention should aim to "transform the relations, norms, and systems that sustain gender inequality and violence" 6 . Of particular interest are gender norms that privilege controlling and aggressive behaviour in the prevailing template for masculinity 6,7 . Efforts to change them are usually termed 'gender transformative' 8 .
A social norm is a belief in the expectations of others in a social group [8][9][10][11] . It is maintained by the influence of a reference group of people important to an individual's decision-making 8,10 . One conceptual norm -a descriptive norm -describes beliefs about what others actually do (roughly equivalent to an empirical expectation or a collective behavioural norm) [12][13][14] . A second -injunctive -norm describes beliefs about what others think one ought to do (equivalent to a normative expectation or a collective attitudinal norm) 12,14 . Cialdini, Kallgren, and Reno contrast these two ideas as what is commonly done and what is commonly approved, or 'is' and 'ought' 12 . An important aspect of a social norm is that it describes what people perceive as the beliefs of the reference group around them, regardless of whether their perception differs from reality. For example, many people in a group may disagree with a norm, but think that others support it. This failure to recognise private disagreement with a perceived norm has been called pluralistic ignorance 9,15,16 , and might at least partly explain why behaviours are sustained when people privately disapprove of them 9 .
Why we do what we do is complicated and norms are only part of the story. For example, a behaviour might be sustained or prevented by social structures such as laws and institutions, material contributors such as wealth (or lack of it), or the availability of services. It might equally be driven by personal beliefs, selfconfidence, and aspirations 17 . From an economic perspective, norms are supported by coordination, social pressure, signalling, and anchoring. Coordination allows individuals to express themselves in shared languages -literally and metaphorically -and benefits both them and the collective. Social pressure encourages individuals not to act purely in their own interest, but that of the collective. Signalling and symbolism allow individuals to identify with (or, equally, indicate their lack of identification with) social groups. Anchoring effectively sets benchmarks for behaviour within a smaller range than what is possible, an example being the ages at which women and men marry 18 . Levy Paluck and colleagues suggest that norms have a stronger influence on an individual's behaviour if they have a clear central tendency (what people do is similar to what they believe others think they ought to do: descriptive and injunctive norms are similar), show little dispersion (such as variation from place to place), and are ascribed to a reference group important to the individual. The greater its importance in her everyday life, the stronger the adherence to a social norm is likely to be 9 .
The resistance of social norms to change varies. They may be sensitive to changes in social networks and to the influence of individuals who emerge as role models or deviants 17 , and their supporting matrix may be complex. For example, violence against women is unlikely to be sustained by a single norm and often occurs at the intersection of gender norms that are permissive rather than supportive 17 . A gender norm is a kind of social norm that describes shared social expectations of behaviour specific to gender 19 . It tends to emerge from gender ideology and attitudes; for example, valuing sons over daughters, idealised conceptions of femininity, and traits that signal masculinity. Glibly, we might think of these as archetypes of the good woman and the real man: constructs that may hinder change rather than actively support violence 20 , but contribute to imbalances of power. These have themselves led to socially constructed gender roles. Most of our discussion is about gender norms that legitimise imbalances of power and, by extension, inequalities in access to resources 19,20 .
Marcus and Harper suggest that gender norms are more likely to change when no parties have strong economic interests at stake, no one's power is directly threatened by change, one key factor underpins a norm, there are no religious injunctions to continue a certain practice, role models and opinion leaders promote changed norms, a changing institutional or political context provides opportunities for changed practices, and norm change communications are paired with opportunities for action 21 . A norm that constrains some people, such as denying education to

Amendments from Version 1
We found the reviewers' suggestions very helpful in pointing out elements that could be added to the introduction and discussion, and in rebalancing the weights of different perspectives on social and gender norms. In reorganising the paper, we have added examples of model interventions in India, relocated our hypothesis to the discussion section, added a discussion of study limitations, added a summary figure, and expanded on our interpretation of the findings. As the reviews point out, the study of norms is complex and multidisciplinary and we have tried to position the implications of our work on a more comprehensive background. We have now included information on the place of social norms within a wider set of influences, the idea of gender norms within gender ideologies and attitudes, economic and psychological perspectives on social norms and changes in them, the notion of in-group and out-group, and a moderation of our discussion of pluralistic ignorance.

See referee reports
REVISED girls, benefits others and it is important to find common ground inasmuch as the perceived net benefit of norm change is positive 20 .
We developed a study to understand them in the context of an urban informal settlement in India, asking three questions: whose opinion matters (the reference group), is the behaviour believed to be typical of the reference group (descriptive norm), and is it believed to be appropriate (injunctive norm)? 8,10 Two-thirds of cities and towns in India include informal settlements (slums) 27 , characterized by overcrowding, insubstantial housing, insufficient water and sanitation, lack of tenure, and hazardous locations 28,29 . There will be over 100 million people in such settlements by 2017 30 , and they currently include 41% of Mumbai's households 27 . Our program has been working in Dharavi, one of Asia's largest informal settlements, since 2000. The population includes a range of cultural and religious groups, whose diversity is illustrated by the availability of schooling in six languages. Some of the 82 geographical clusters are homogeneous, retaining cultural group identity. Others are heterogeneous, but all embody the idea of villages within the city 31 . This, and factors like poverty, poor housing and environment, and governmental neglect, make Dharavi a backdrop for the operation of identity politics which are predominantly communal and patriarchal.

Study procedures
Before data collection, we convened a group of fieldworkers and counsellors to develop a provisional set of examples of social norms, using anonymised existing records from 1653 clients registered at our crisis and counselling centre in 2012-2015.
Counsellors then selected cases purposefully to reflect a range of presenting problems (for example, natal family violence, intimate partner or domestic violence). This preliminary exercise reached saturation after detailed discussion of 25 cases (Table 1). While most of the examples were social norms, a few -"boys will be boys," "women should not have sex before marriage," and "it is a good woman's duty to make her husband's family happy" -had the character of more general moral norms. Having identified problems presented by age, religion, caste, employment, and wealth, we developed a fictional vignette for discussion, covering premarital life, getting married, and marital life.
Responses to survey questions on gender norms -such as those used in Demographic and Health Surveys -may differ from responses to more contextualised questions based on illustrative vignettes 32 . Over three months in 2016, we held a series of focus group discussions and individual interviews. A purposive sample of female and male participants represented two age groups (18-30 and 30-55 years) and four localities. We intentionally included more women than men because our preventive program works primarily with women as drivers of norm change. A team of ten went door-to-door to recruit participants from pre-selected regional, religious, or cultural communities, mobilising women by age group as participants for focus group discussions. Simultaneously, counsellors referred clients by age group and community. Three postgraduate female researchers, already working with the program and with experience of qualitative research, conducted focus group discussions and interviews with women. Men were interviewed by a graduate male community worker. Because we wanted to minimise social desirability bias as a result of previous exposure to program activities, we invited people unfamiliar with our work to participate. Participants and researchers generally met for the first time at discussions and interviews. The researchers talked about their professional backgrounds, experience, and the reasons for the study. They described violence as potentially affecting all women, including themselves, and urged participants to speak candidly so that their opinions could be used to design interventions that would help others. They assured participants that they could contact a supervisor if they had concerns.
Focus group discussions were held in community spaces familiar to participants: a program community centre, the homes of community volunteers or participants, and a temple. Interviews with women were conducted at a municipal Urban Health Centre. Women often brought children and grandchildren to discussions and the researchers provided drawing materials to occupy them. Women who did not fit the focus group age bracket were often present.
The vignette used in focus group discussions followed a hypothetical biography as a means of eliciting opinions 8 , and was piloted with a mixed group of male and female staff. It traced a woman's life from later schooldays until about ten years into her marriage. Example scenarios included having a boyfriend whom her parents did not know about, her parents checking her mobile phone, not wanting to have children soon after her wedding (when her in-laws wanted her to), wanting in-laws to take care of her first child so that she could complete her education, her husband realising she had male friends at college, her desire to give some of her salary to her parents, being delayed at work and not preparing food for her husband and children, her husband suspecting that she was communicating with another man by smartphone, and discovering that her husband was having an affair. A child may be used as leverage over a woman after she has left an abusive situation. It is acceptable for in-laws to retain custody of a male (grand)child after a woman separates from her husband.
Police Police see domestic violence as a private matter and do not necessarily take it seriously. Police are reticent to take action that is challenging. It is acceptable for the police to call a woman's parents, irrespective of her views.
Beyond the family Community leaders have a say in issues such as marriage.
Interviews followed a semi-structured topic guide in which participants were asked to give a biographical account of their lives. Discussions and interviews were audio-recorded and researchers took notes to triangulate transcription and translation. Transcripts were translated from Hindi and Marathi to English by KH and an independent translator, with review for accuracy by PP. The average duration of focus group discussions was 53 minutes, and of interviews 44 minutes. Data saturation was discussed when designing the study and during the course of data collection. Participants were given the option of reviewing the recordings or transcripts of their interviews, but none requested to do so.

Ethical statement
The study was approved by the 'Ethicos' Independent Ethics Committee, Mumbai, on 3rd December 2015. We reviewed our existing case records to identify norms for consideration. Signed informed consent is taken from all clients when they first access our services. Counselors inform them of their right to access their records in building evidence for their legal cases. They are told that their anonymised information may be used in research conducted by the organisation, with the aim of improving services. Clients are assured of confidentiality, particularly that information will not be shared with the perpetrator's family, community members, or the media.
Participant information sheets and written consent forms (in English and Hindi, read aloud when appropriate) were given to participants before interviews and focus group discussions, and all participants provided signed consent. Although the interviews and focus groups did not consider participants' own experiences, interviewers were familiar with existing organisational response and referral protocols and followed WHO recommendations for research on violence against women 33 . Participants were advised that they should consider as confidential any information shared by other participants in focus groups that might have referred to local individuals.

Analysis
We developed a framework analysis, beginning with a provisional set of norms, attitudes and beliefs identified from the literature, case review, and practitioner workshop. We used framework analysis because we came to the study with a provisional classification of norms, and had an agreed sample and timeline 34,35 . Transcripts were analysed in NVivo 10 (www.qsrinternational. com). Because focus groups and interviews followed a semistructured sequence, we began with a list of general coding categories, which we expanded and sub-categorised into a coding tree in a series of team discussions 36-39 . These categories described types of gender norm, response to transgression, and classification as descriptive or injunctive. We revisited the analysis repeatedly over six months, reviewing individual transcripts and hierarchies of categorical codes in an effort to achieve a higher level of thematic description. We settled on the comparison of descriptive and inductive norms early, but the idea of the importance of reference groups emerged much later.

Results
Areas in which injunctive and descriptive norms coincided: femininity, masculinity, marriage, childbearing, separation, and friendship across sexes Table 2 summarises discussions with 56 women and 15 men between 30 th September and 23 rd December, 2015 (10 interviews and 8 focus group discussions), which supported views of women's and men's roles familiar from our work and the literature. Older men and women favoured arranged marriages, a belief in which descriptive and injunctive norms coincided. They said that a woman became the responsibility of her in-laws after marriage and that her mobility, even to her natal home, needed to be monitored. While intimate partner and domestic violence were disapproved of, most participants said that some violence was acceptable if a woman neglected the house and children, did not cook properly, was unfaithful, or disrespected her in-laws. Shouting and slapping were acceptable, but anything more was not. Older men said that, as household heads, their masculinity depended on their roles within and outside the home.
"Menhavetoearn,takecareofthefamily,keepthepeople aroundorganizedandunited.Alazymanwon'tbeacceptedby societyandpeoplewilltaunthim." Interview, older Hindu man (OM-I-89) Masculine behaviour was upheld by social sanctions. Younger men and women agreed that there was nothing wrong with a man cooking or helping his wife with domestic chores, but men refrained from doing so because they thought they would be gossiped about and taunted. Women agreed that men preferred not to help because they were conscious of their peers and families impugning their masculinity. Participants often defined masculine behaviour against the construct of the good woman, for whom the level of sanction was more punitive: participants in focus group discussions said that it was acceptable to beat a woman if she did not conform to the archetype of a good daughter, wife, mother, or daughter-in-law. Injunctive and descriptive norms coincided in the belief that women should get married and have children, irrespective of their education and income and whether or not they married willingly. In one community, adolescent marriage was still common. Mothers were not keen to relax this expectation, even though it was somewhat at odds with their belief that girls should be educated: Interview, younger Hindu woman (YW-I-12) Women said that they, or people they knew, often stayed in abusive marriages for fear of the implications of separation. They had seen or heard of divorced women being gossiped about, taunted, and abused. This was a strong reason for staying, apart from factors such as financial dependence on the perpetrator, lack of confidence, and worry about the children. A woman who separated from her husband was expected not to talk to men unknown to the family, not to entertain proposals from men, and to cater to her children's needs above all else. Both injunctive and descriptive norms were for her not to divorce. If she did, the injunctive norm was for her to return to her natal family, who might or might not support her decision. The commonest sanction for transgression was reputational damage.

FGD, younger women (YW-F-52)
Women almost all said that they did not have male friends when they were young for fear of retribution from male relatives. They said that men were quite comfortable with beating female relatives for transgression. Interview, younger Hindu woman (YW-I-11) During a discussion on premarital sex, older women said that girls faced violence because their actions were wrong and boys were often beaten because they had tainted the girl's family honour.

FGD, older women (OW-F-51)
Some women interviewees said that they were aware of, or were suspicious of, their husbands having sex outside marriage and seemed to accept it. The same women, however, had approached our organisation because their husbands had physically abused them when they thought they were talking to other men. Only one woman said that she had had an extramarital relationship. The acceptance with which wives treated their husbands' behaviour did not extend to her. Control of sexual activity included sanctions against women using mobile phones. Older women said that only women who were "up to no good" needed "advanced phones" because they used them to exchange messages and photos with other men. Interviewees who had been in abusive marriages said that they would be physically and verbally abused for using a mobile phone.
Interview, younger Hindu woman (YW-I-12) Areas in which injunctive and descriptive norms differed slightly: dress, mobility, and visibility Older men said that women from their area who dressed inappropriately invited sexually coloured remarks and that good women dressed appropriately. In some cases, however, this attitude was relaxed toward women perceived as higher-class.

Interview, younger Muslim man (YM-I-14)
Areas in which injunctive and descriptive norms differed markedly: education, earnings, premarital relationships, and premarital sex As with employment, there was a clear impression of change in norms around female education. Participants were unanimous in their belief that the practice of not educating girls had to change. They also supported the idea of continuing education after marriage and childbirth.

FGD, younger women (YW-F-20)
This commitment to women's self-determination wavered when the demands of education compromised their ability to fulfil gendered family responsibilities.
Men reported relatively progressive attitudes to gender roles in decision-making. Although older men stressed that husbands should be providers, they said that wives should have control over their own earnings. A woman could decide what to do with income: either contribute to household expenditure or save for an emergency. Emergencies were themselves gendered, relating to the woman's reproductive role, examples being children's illnesses or demands for school fees. Although participants mentioned that women's expenditure had traditionally been controlled by their families, the idea of a woman earning her own money was nothing special. There were, however, limits: interviewees found the idea of a woman spending her entire income on her own needs selfish. The 'traditional' norm was for the natal family to have no call on a woman's income once she entered her in-laws' home, but younger men and women said that she had the right to choose whether she wanted to give part of it to her own parents.

FGD, younger Muslim woman YW-I-92
The idea that young people should take seriously the advice of their elders on relationships and sex was more of an injunctive than a descriptive norm, and it happened rarely. Although not entirely comfortable with premarital sex, older men and women seemed to be resigned to the fact that young girls and boys were sexually active and were making sexual decisions irrespective of their marital status.

FGD, older women (OW-F-51)
Women's sexuality was bound up with family status. The primary responsibility for premarital sex fell on girls and women because of fear that an unwanted pregnancy could dishonour the family.
The importance of reference groups Across age groups, women and men described their reference groups as 'society' or 'community'. What these words meant was less clear, as we have found in other research 40 . Women interviewees often talked about society as an abstract body that dictated their decisions, mobility, and responsibilities, but struggled to define it further. Although these words call to mind large collectives, their interpretations were actually quite limited. For women, the community was the family. Men seemed to have a broader conception of community and society, possibly because of their mobility and social interactions. Hindu and Muslim women said that they respected the authority of elders from the Panchayat and the Jammat, respectively, and men talked about performing gender roles according to the norms dictated by religious bodies. Older men said that their behaviour was often shaped by the opinions of those in power within these local organisations. There were gendered differences in the behaviour of such bodies. Younger men were more cynical about traditional authorities and their influence on gender roles and ideologies.

FGD, younger men (YM-I-14)
Women's behaviour and decisions were dictated more by their families -natal families before marriage and affinal afterwards -and by the notion of reputation among neighbours. Interview, younger Hindu woman (YW-I-11) Women often took pride in the fact that they did not interact with people outside their families, a trait that made them good women in family eyes. This extended to abandoning friends, particularly in the case of male friends after marriage, which further reduced their networks of interaction and narrowed their reference group. Interview, younger Hindu woman (YW-I-26) We take two ideas forward from these illustrations. First, reference groups were tight and generally involved strong social ties (families, local cultural and religious organisations) 41 , which we might think of in terms of bonding social capital 42 . For women, they were extremely tight and often confined to close family. Men described wider networks, albeit still fairly local. These reference groups were conduits for the opinions of a wider reference group, locally in the form of neighbours and distantly in something called society. Our inference is that these closely bonded reference groups were likely to transmit -and transmute -injunctive norms from a dimly perceived wider world.

Discussion
Focus group discussions and interviews involving 71 women and men in an urban informal settlement in Mumbai suggested that injunctive and descriptive norms were relatively similar with regard to femininity, masculinity, the need for marriage and childbearing, resistance to separation and divorce, and disapproval of friendships between women and men. Some constraints on women's dress and mobility were relaxing, but there were more substantial differences between injunctive and descriptive norms around women's education, control of income and finances, and premarital sexual relationships (Figure 1).
Why should this be so? It seems to us that the areas of greater discrepancy -loosely, where norms might be seen as weaker -reflect both societal change and resource constraint. Many precedents, campaigns, and programs encourage female education, and this is augmented by the observation that, in urban Mumbai at least, girls nowadays usually go to school. In this sense, the descriptive norm favours female education -up to a point -and perhaps the injunctive norm is itself changing. Urban life is expensive and people are time-poor. Women need to be able to manage household expenses and income from their own employment. At the same time, the gendered division between reproductive and productive work has historically meant that domestic parsimony and responsibility are characteristics of a good wife and mother. Whether or not there is increasing permissiveness toward premarital sex, the penalties for young women can be severe. Young men, however, have traditionally been granted the privilege of premarital sexual relationships and, although this licence may have more of a passive than an active character, it is arguably part of their gender role.
There were some limitations to our study. We purposefully interviewed more women than men and we chose an informal settlement which -though very large and occupied by communities of diverse origin -may limit generalisation. The use of vignettes might also have led to a focus on the scenarios that they included, to the exclusion of other possible issues. We have some assurance, however, from the extensive dataset of client records on which the vignettes were based.
Recent reviews have identified community mobilisation as a potentially effective means of reducing violence against women and girls 43 . There are some models for programs 44-50 , all of which aim to change norms, but the evidence for effectiveness is not yet robust 6 . Several programs have worked to change norms in India. The Samata program seeks to reduce HIV acquisition and child marriage through changing education norms for lower caste girls in Karnataka state. Also in Karnataka, Samvedana Plus addresses violence against sex-working women 51,52 . The Yari-dosti program promotes gender equity among young men in low-income communities in Mumbai 53 , and PRACHAR sought to improve young people's sexual and reproductive health in Bihar 54 .
A report on norms around violence against women and girls in rural Bihar 55 , framed its findings around the 'real man' and 'real woman' and the 'good' or 'bad' husband. A real man fulfilled the responsibilities of breadwinner, household head, son, husband, and father. He also made his wife happy, respected her views, did not impose restrictions on her, and -importantly -did not perpetrate violence. A real woman was nurturing, caring for home and children and serving and obeying her husband and his parents. A good husband managed household income responsibly and treated his wife with respect, did not abuse alcohol, and was not violent. A bad husband squandered money, abused alcohol, and perpetrated violence, particularly sexual violence. Perhaps the most interesting aspect of the findings for our work was that intimate partner and domestic violence were generally disapproved of: the injunctive norm was non-violence. Violence was, however, acceptable in some situations -such as infidelity -which might be interpreted differently depending on the perspectives of perpetrator and survivor. Recent evidence suggests that this loophole might be wider if transgression is perceived to be intentional rather than unintentional 32 .
Programmatic response 1: exploit the mismatch between injunctive and descriptive norms Norms are characterized by persistence, punctuated equilibrium and tipping points, compression (the range of individual choices varies less than one might expect), and local conformity within global diversity 18 . Developing our idea about the utility of dissonance between descriptive and injunctive norms, we think that the idea of building a new norm might be less useful than supporting the emergence of a norm that is already developing. Our hypothesis was that we might exploit the disjunction between descriptive and injunctive norms, focusing program activities on evidence that descriptive norms are changing. In doing so, we might also discuss the relatively weaker sanctions for transgression in this situation. One way to do this might be to identify norms for which disjunction already exists, and then to emphasize the numbers of people who are transgressing: to shift the gaze to a source of information that supports change in descriptive norms 12 .
There is implicit support for this in the literature: Heise and Manji suggest "communicating change as norms begin to shift." 11 The observation that "norm change is particularly likely in homogenous, tightly knit groups in which there is private dissent against the current norm" 9 is also interesting because it seems to imply at least dissonance between injunctive and descriptive norms, if not actual changes in expectations. In a gaming experiment, Bicchieri and Xiao suggested that the driver of conformity with a norm was descriptive: what the individual believed others actually did. Injunctive norms seemed to play a part only when they were in line with this belief 56 . One possible explanation for this is that punishment for perceived transgression is less likely when many people are transgressing: "Descriptive norms act like magnets, whereas injunctive norms act like bans." 9 From a psychological perspective, emphasizing a descriptive norm that differs substantially from an injunctive norm could be expressed as 'personalised normative feedback 57 ,' in which people are shown the discrepancies between their estimates of what is usual and what is actually usual.
Programmatic response 2: expand the reference group There seems to be general agreement that interventions should target injunctive norms, which "… make it clear to all members of the community that the particular behavior is not welcome." 9 Perhaps counter-intuitively, a good way to do this may be to use evidence of changes in descriptive norms because "… enough members of the group must believe that enough of its members are adopting the new norm." 10 Our program presents us with an environment in which we can support norm change by providing conduits through which individuals can articulate it. These 'channel factors' 9 -in our case, community groups, model change agents, and the availability of counselling, family therapy, and legal support -make it easier for individuals to change by providing structure and support 58 .
We have noticed that discussions of the means to change norms have said little about changing the reference group. Since norms are maintained by sanctions within the reference group 8 , the idea is to help cement new shared beliefs in the key people from whom the individual takes her cue 8,11 . Through no fault of their own, the women we interviewed and, to a lesser degree, the men, looked to small reference groups for guidance. The kinds of interest groups whose activities we facilitate have their own sets of injunctive and descriptive norms and, at least to some degree, either erode the salience of existing reference groups or replace them. A corollary of group behaviour is that an individual can use its symbolic features to either identify herself as a member (in-group) or to identify a group as an entity of which she is decidedly not a member (out-group). Efforts to change norms may therefore look to participants distancing themselves from groups that maintain behaviours seen as unjust, or shifting their identification to groups with more conducive norms 59 . We would argue that the ubiquity of action groups, communities of interest, and identity politics implies that individuals are able to find safe harbour in a reference group that they have chosen as reflecting their own aspirations and differing from those of others. This idea of expanding the reference group has wide applicability and underlies, for example, participatory programs to improve child survival 60 , nutrition 61 , and adolescent sexual and reproductive health in India and the region 54 , all of which are affected by normative behaviour.

Conclusions
Our findings have crystallised our thinking about norms in light of 16 years of work to address violence against women and girls in informal settlements in India. We would like to suggest a different way of looking at norm change, on the basis of three propositions. First, as already suggested, we should take advantage of the mismatch between descriptive and injunctive norms, given that descriptive norms intolerant of violence are likely to be magnets for behaviour and that sanctions will become less aggressive as adoption increases.
Second, we should take advantage of the existence of injunctive norms that are already inimical to violence against women and girls. We see this articulated by our interviewees and in population surveys 62 , but disapproval is already explicit in concentric reference groups that take in the state, the media, and the wider world. The socio-ecologic model tends to be invoked in a negative sense, to emphasise the need for change at multiple levels. It also has positive implications. Societal changes such as new legislation and global influences combine with education, incremental improvements in economic status, and exigencies such as the need to work, to create a milieu favourable to change.
What we are seeing across the world -with some exceptions -is gender equity taking on the status of a high-level injunctive norm. This may not be articulated at the level of an individual's reference group (indeed, this is part of the problem), but it filters down through the strata of the socio-ecologic model and can provide a platform for mobilisation. At the same time, increasing education, mobility, and employment are expected to lead to more friendships outside the family circle, and potentially to exposure to wider cultural changes.
Third, we suggest that the key intervention is not only to make people aware of this, but, in doing so, to induct them into a wider or different reference group. This is central to what community interventions actually do: expand participants' social world to encompass others whose opinions might differ from those they have been exposed to. This expansion of the reference group takes us into an area of considerable experience and theory that has not yet been reflected substantially in the literature on violence against women and girls: the idea that social capital and, particularly, bridging rather than bonding social capital, or weak rather than strong ties, might be associated with wellbeing. We suggest, therefore, that programs to prevent violence focus on creating new ties for women and girls whose social environments are limited. Further support for this comes from anthropological ideas about mitigating violence through intra-group linkages 63 . This is already implicit in much of our work, which involves forming and facilitating groups. What has not been explicit is the importance of expanding the reference group to one that has already abandoned harmful injunctive norms and is in the process of challenging them through new descriptive norms.

Data availability
Transcripts of focus group discussions and interviews, translated into English, are available at the UK Data Service ReShare (http://reshare.ukdataservice.ac.uk/), under the title 'Changing gender norms in the prevention of violence against women and girls in India', DOI, 10.5255/UKDA-SN-852735. The safeguarded data files are made available to users registered with the UK Data Service under UK Data Archive End User Licence conditions (https://www.ukdataservice.ac.uk/get-data/how-to-access/ registration). The data files are not personal, but -given the subject matter of the interviews and focus groups -the data owner and research ethics committee consider there to be a limited residual risk of disclosure.

Competing interests
No competing interests were disclosed.

2.
3. This is a powerful article that touches upon the extremely important intersection between intimate partner violence and social norms. Overall this is an interesting article and offers some important insights for practitioners and researchers working at the intersection between social psychology and gender theory. I highlight its three main merits and shortcomings below, and offer a more detailed review in the second part of this comment.
Three merits of this paper: Is located at the intersection between gender theory and social psychology, working on an important conceptual no people's land that needs to be further explored.
Introduces the reader to the important ways in which descriptive and injunctive norms can interact and how their asynchrony can offer opportunities for social change.
Conducts an interesting social norms analysis from qualitative data, on the line of previous similar work in this field, that can be used as model for others.
Three shortcomings of this paper: It presents the importance of doing social norms work as tightly related to pluralistic ignorance (a phenomenon occurring when most people's attitudes are opposite to what they believe others' attitudes to be). But norms and attitudes can be aligned, and yet achieving change requires to address both norms and attitudes.
It invites the reader to analyse gendered relations and practices through a social norms lens, but doesn't integrate a thorough conceptual understanding of the difference between gender norms and social norms. A larger discussion of how gender (roles, norms, schemas, identity) differ would also be helpful.

Introduction
Social norms literature is vast and multifaceted; the authors decided to use the Focus Theory of Normative Conduct developed by Cialdini and colleagues. Their conceptual understanding of social norms could expand to include awareness that other reasons for compliance exist other than normative influence (what the authors call "the influence of a reference group of people important to an individual's decision-making", with a phrase that echoes the Theory of Planned Behaviour). The most important of these mechanisms for compliance being: 1) social categorisation, 2) social identity, 3) informational influence (coordination and cooperation); and 4) power . The authors could also expand on the strength of the norms they identified. As myself and Heise argued elsewhere , there is a tendency in the world of global justice and international development to look at social norms as on/off switches, overlooking their . The authors make great use of the concept of "pluralistic ignorance", when norms and individual attitudes are not aligned. This is an insightful concept, but it's important to stress that the opposite can also (and quite often is) true: social norms and individual attitudes can be aligned; norms can actually shape those attitudes (as scholars working in the gender theory field might argue).
The authors use gender and social norms almost interchangeably; it would be important to understand if they think they differ at all. The social norms construct they use comes from a social psychology, but much theory around gender norms has been developed by scholars conceptually distant from Cialdini and colleagues. The authors mention that examples of community mobilisations are rare; they do mention some of these examples and could also mention well-known cases . In addition, they could look at the case of social norms and intimate partner violence in Southern India discussed in Cislaghi & Bhattacharjee 2017 .

Methods
Study procedures are very thoroughly explained and well thought. Sometimes, however, the "provisional normative statements" included in Table 1 leave the doubt on whether those statements are all socially normative or morally normative. The difference is sometimes subtle and understandings can overlap.

Results
The results offer a solid qualitative social norms analysis. Authors could have mapped the descriptive and injunctive norms in a table, to show their concordance or discordance. Sometimes it is not completely clear whether authors are looking at norms as defined in the social psychology literature they cite in the intro, or more in the way they have been articulated in applied gender work (for instance at p.8).

Discussion
Some of these findings resonate with a similar study conducted in Southern India and that would be probably be an interesting read for the authors (8). The discussion brings in gender norms as a construct but -as mentioned earlier -more understanding of what gender norms actually are (and how they relate the Cialdini's social norms used by the authors) would be helpful-

Conclusions
An understanding of the difference between outgroup and ingroup is missing. That is dangerous, because (as the work by Kees Keiser shows) implementing interventions that do not target the right group can increase compliance with the harmful norms with the outgroups ("that message is not for me -I'd better do X even more to make sure people know I am not one of those"). The intervention speaks back to correcting misperceptions, but literature on violence shows that needs to go with local discussions on the nature of harm and protection, because attitudes and norms might be synchronous. As mentioned earlier, the conclusions overstate the importance of pluralistic ignorance: evidence from West Africa suggests that dialogue can increase the mismatch between attitudes and norms or descriptive and injunctive norms required to bring change at scale. . 2017.

London: STRIVE
Reference Source

If applicable, is the statistical analysis and its interpretation appropriate? Not applicable
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Yes
No competing interests were disclosed.

Competing Interests:
I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
Author Response 09 Nov 2017 , University College London, UK

David Osrin
Thank you for the useful and incisive review. A summary of our responses follows. This is a powerful article that touches upon the extremely important intersection between intimate partner violence and social norms. Overall this is an interesting article and offers some important insights for practitioners and researchers working at the intersection between social psychology and gender theory. I highlight its three main merits and shortcomings below, and offer a more detailed review in the second part of this comment. Three merits of this paper: 1. Is located at the intersection between gender theory and social psychology, working on an 1.
1. Is located at the intersection between gender theory and social psychology, working on an important conceptual no people's land that needs to be further explored. 2. Introduces the reader to the important ways in which descriptive and injunctive norms can interact and how their asynchrony can offer opportunities for social change. 3. Conducts an interesting social norms analysis from qualitative data, on the line of previous similar work in this field, that can be used as model for others. Three shortcomings of this paper: It presents the importance of doing social norms work as tightly related to pluralistic ignorance (a phenomenon occurring when most people's attitudes are opposite to what they believe others' attitudes to be). But norms and attitudes can be aligned, and yet achieving change requires to address both norms and attitudes. It invites the reader to analyse gendered relations and practices through a social norms lens, but doesn't integrate a thorough conceptual understanding of the difference between gender norms and social norms. A larger discussion of how gender (roles, norms, schemas, identity) differ would also be helpful. It underplays the role that self-understanding as in-or out-group plays in influencing people's decision to comply with a norm. Detailed feedback: This is an important paper and the work is interesting. The work on social norms and violence is still emerging as a field of research and action, and much is left to learn on how the theory can best serve the practice. For this reason, papers such as this one are very welcome. Introduction Social norms literature is vast and multifaceted; the authors decided to use the Focus Theory of Normative Conduct developed by Cialdini and colleagues. Their conceptual understanding of social norms could expand to include awareness that other reasons for compliance exist other than normative influence (what the authors call "the influence of a reference group of people important to an individual's decision-making", with a phrase that echoes the Theory of Planned Behaviour). The most important of these mechanisms for compliance being: 1) social categorisation, 2) social identity, 3) informational influence (coordination and cooperation); and 4) power . The authors could also expand on the strength of the norms they identified. As myself and Heise argued elsewhere , there is a tendency in the world of global justice and international development to look at social norms as on/off switches, overlooking their varying strength, a point also mentioned in Cislagi & Heise 2016and [report]. The authors make great use of the concept of "pluralistic ignorance", when norms and individual attitudes are not aligned. This is an insightful concept, but it's important to stress that the opposite can also (and quite often is) true: social norms and individual attitudes can be aligned; norms can actually shape those attitudes (as scholars working in the gender theory field might argue). Response Thanks for these excellent criticisms. Responding to them underlies many of the changes we have made to the paper. We have added to the introduction some text on reasons for compliance with norms: "Why we do what we do is complicated and norms are only part of the story. For example, a behaviour might be sustained or prevented by social structures such as laws and institutions, material contributors such as wealth (or lack of it), or the availability of services. It might equally be driven by personal beliefs, self-confidence, and aspirations. From an economic perspective, norms are supported by coordination, social pressure, signalling, and anchoring. Coordination allows individuals to express themselves in shared languages -literally and metaphorically -and benefits both them and the collective. Social pressure encourages individuals not to act purely in their own interest, but that of the collective. Signalling and symbolism allow individuals to identify with (or, equally, indicate their lack of identification with) social groups. Anchoring effectively sets 1 2 equally, indicate their lack of identification with) social groups. Anchoring effectively sets benchmarks for behaviour within a smaller range than what is possible, an example being the ages at which women and men marry." We have added a mention of the strength of norms we identified in the discussion: "Why should this be so? It seems to us that the areas of greater discrepancy -loosely, where norms might be seen as weaker -reflect both societal change and resource constraint. Many precedents, campaigns, and programs encourage female education, and this is augmented by the observation that, in urban Mumbai at least, girls nowadays usually go to school. In this sense, the descriptive norm favours female education -up to a point -and perhaps the injunctive norm is itself changing. Urban life is expensive and people are time-poor. Women need to be able to manage household expenses and income from their own employment. At the same time, the gendered division between reproductive and productive work has historically meant that domestic parsimony and responsibility are characteristics of a good wife and mother. Whether or not there is increasing permissiveness toward premarital sex, the penalties for young women can be severe. Young men, however, have traditionally been granted the privilege of premarital sexual relationships and, although this licence may have more of a passive than an active character, it is arguably part of their gender role." We have also tried to make pluralistic ignorance less of a feature of the paper and it is now mentioned on once. The authors use gender and social norms almost interchangeably; it would be important to understand if they think they differ at all. The social norms construct they use comes from a social psychology, but much theory around gender norms has been developed by scholars conceptually distant from Cialdini and colleagues. The authors mention that examples of community mobilisations are rare; they do mention some of these examples and could also mention well-known cases. In addition, they could look at the case of social norms and intimate partner violence in Southern India discussed in Cislaghi & Bhattacharjee 2017. Response Thanks again. We have added a clarification on gender norms to the introduction: "The resistance of social norms to change varies. They may be sensitive to changes in social networks and to the influence of individuals who emerge as role models or deviants, and their supporting matrix may be complex. For example, violence against women is unlikely to be sustained by a single norm and often occurs at the intersection of gender norms that are permissive rather than supportive. A gender norm is a kind of social norm that describes shared social expectations of behaviour specific to gender. It tends to emerge from gender ideology and attitudes that have themselves led to socially constructed gender roles; for example, valuing sons over daughters, idealised conceptions of femininity, and traits that signal masculinity. Most of our discussion is about gender norms that legitimise imbalances of power and, by extension, inequalities in access to resources. Glibly, we might think of these as archetypes of the good woman and the real man: constructs that may hinder change rather than actively support violence, but contribute to imbalances of power." We have included other sources of scholarship in the introduction… "From an economic perspective, norms are supported by coordination, social pressure, signalling, and anchoring. Coordination allows individuals to express themselves in shared languagesliterally and metaphorically -and benefits both them and the collective. Social pressure encourages individuals not to act purely in their own interest, but that of the collective. Signalling and symbolism allow individuals to identify with (or, equally, indicate their lack of identification with) social groups. Anchoring effectively sets benchmarks for behaviour within a smaller range than what is possible, an example being the ages at which women and men marry." … and in the discussion: Methods Study procedures are very thoroughly explained and well thought. Sometimes, however, the "provisional normative statements" included in Table 1 leave the doubt on whether those statements are all socially normative or morally normative. The difference is sometimes subtle and understandings can overlap.

Response
We have added to the methods section: "While most of the examples were gender norms, a few -"boys will be boys," "women should not have sex before marriage," and "it is a good woman's duty to make her husband's family happy"had the character of more general moral norms."

Results
The results offer a solid qualitative social norms analysis. Authors could have mapped the descriptive and injunctive norms in a table, to show their concordance or discordance. Sometimes it is not completely clear whether authors are looking at norms as defined in the social psychology literature they cite in the intro, or more in the way they have been articulated in applied gender work (for instance at p.8). Response Thanks. We made a table, but then thought that a summary diagram might be nicer, and added Figure 1. Discussion Some of these findings resonate with a similar study conducted in Southern India and that would probably be an interesting read for the authors (8). The discussion brings in gender norms as a construct but -as mentioned earlier -more understanding of what gender norms actually are (and how they relate the Cialdini's social norms used by the authors) would be helpful.

Response
We hope that the revisions we have mentioned previously are an adequate response to this comment. Conclusions An understanding of the difference between outgroup and ingroup is missing. That is dangerous, because (as the work by Kees Keiser shows) implementing interventions that do not target the right group can increase compliance with the harmful norms with the outgroups ("that message is not for me -I'd better do X even more to make sure people know I am not one of those"). The intervention speaks back to correcting misperceptions, but literature on violence shows that needs to go with local discussions on the nature of harm and protection, because attitudes and norms might be synchronous. As mentioned earlier, the conclusions overstate the importance of pluralistic ignorance: evidence from West Africa suggests that dialogue can increase the mismatch pluralistic ignorance: evidence from West Africa suggests that dialogue can increase the mismatch between attitudes and norms or descriptive and injunctive norms required to bring change at scale. Response As mentioned before, we have cut down the mention of pluralistic ignorance. We have also added to the discussion in-group and out-group issues, which we found a particularly interesting insight from the reviewer: "A corollary of group behaviour is that an individual can use its symbolic features to either identify herself as a member (in-group) or to identify a group as an entity of which she is decidedly not a member (out-group). Efforts to change norms may therefore look to participants distancing themselves from groups that maintain behaviours seen as unjust, or shifting their identification to groups with more conducive norms. We would argue that the ubiquity of action groups, communities of interest, and identity politics implies that individuals are able to find safe harbour in a reference group that they have chosen as reflecting their own aspirations and differing from those of others."

Introduction
The authors provide a comprehensive introduction into the role of social norms, particularly gender norms and the social construction of masculinity, in violence against women and girls. While the authors argue that the evidence for effectiveness of community mobilization interventions is "not yet robust," I believe it would strengthen the paper to add 1-2 examples of such interventions in urban India. The models for programs that the authors cite are great examples, all but one are from outside India. Though there are numerous challenges to evaluating such interventions, there are a number of community mobilization interventions in India in the literature (to name a few, there have been evaluations from ICRW, Population Council, UNFPA).
While the study questions are well conceptualized. The hypothesis is actually not a hypothesis for this study, since the study itself is not on an intervention exploiting the disjunction between descriptive and injunctive norms. It is more of a future interest of how the study findings could be applied. I think that a hypothesis is not strictly necessary due to the exploratory qualitative nature of the study. The hypothesis could then be reframed as an interest of the study. The second half of the first paragraph in the second column of page 3 could be moved into the discussion, since it is more on how the results can be applied.

Methods
The authors provide a clear description of the study community. It is clear that the organization has the unique perspective of having strong contextual knowledge of the study population. The authors also have a unique dataset of clients at the crisis and counseling center, generating a comprehensive list of normative statements to inform the qualitative data collection. normative statements to inform the qualitative data collection.
It would be great if there was a small statement on the possible limitations of using vignettes, in that it leads participants to discuss certain scenarios, potentially limiting the topics covered in the discussion. However, we understand that this could be mitigated by using the dataset of clients at the counseling center. It is very much appreciated that you have published your vignette script, it is a great service to other organizations conducting this type of research in India.
It would also have been great to include a bit more information included on the Framework Analysis approach and how it was applied.

Results
It is not discussed why there more women recruited than men (56 women and only 15 men). It is unclear as to whether this was intentional or not? If intentional, that should be stated. If not, it would be great to have some discussion on the challenges of recruiting men for the study.
We would also be careful with comparing Hindu and Muslim women in the second column of page 8 since the sample is limited to 5 women in the in-depth interviews for each.

Discussion and Conclusion
It would be great if the authors could provide some thoughts on why there were more substantial differences between injunctive and descriptive norms around women's education, control of income and finances, and premarital sexual relationships, but not in the other areas covered in this study. It also seems like the results from the study in Bihar are not integrated into the rest of the discussion. It does not flow with the paragraph before and after it.
The discussion and conclusion provide a great application of the results to the work that is being done by the organization. It would be great if the authors could also add how these results could also be applied to other work being done in India, perhaps citing a few other types of interventions that could also benefit from the findings of this study.

If applicable, is the statistical analysis and its interpretation appropriate? Not applicable
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Yes
No competing interests were disclosed.

Competing Interests:
I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard.
a great service to other organizations conducting this type of research in India.

Response
We have mentioned this in the discussion in a new paragraph on limitations of the study. "There were some limitations to our study. We purposefully interviewed more women than men and we chose an informal settlement which -though very large and occupied by communities of diverse origin -may limit generalisation. The use of vignettes might also have led to a focus on the scenarios that they included, to the exclusion of other possible issues. We have some assurance, however, from the extensive dataset of client records on which the vignettes were based." It would also have been great to include a bit more information included on the Framework Analysis approach and how it was applied.

Response
We have added to the paragraph in the methods: "Because focus groups and interviews followed a semi-structured sequence, we began with a list of general coding categories, which we expanded and sub-categorised into a coding tree in a series of team discussions. These categories described types of gender norm, response to transgression, and classification as descriptive or injunctive. We revisited the analysis repeatedly over six months, reviewing individual transcripts and hierarchies of categorical codes in an effort to achieve a higher level of thematic description. We settled on the comparison of descriptive and inductive norms early, but the idea of the importance of reference groups emerged much later." Results It is not discussed why there more women recruited than men (56 women and only 15 men). It is unclear as to whether this was intentional or not? If intentional, that should be stated. If not, it would be great to have some discussion on the challenges of recruiting men for the study. Response It was intentional and we have added to the methods paragraph: "A purposive sample of female and male participants represented two age groups (18-30 and 30-55 years) and four localities. We intentionally included more women than men because our preventive program works primarily with women as drivers of norm change." We would also be careful with comparing Hindu and Muslim women in the second column of page 8 since the sample is limited to 5 women in the in-depth interviews for each.

Response
We agree and have removed mention of this. Discussion and Conclusion It would be great if the authors could provide some thoughts on why there were more substantial differences between injunctive and descriptive norms around women's education, control of income and finances, and premarital sexual relationships, but not in the other areas covered in this study.

Response
We have added to the early part of the discussion: "Why should this be so? It seems to us that the areas of greater discrepancy -loosely, where norms might be seen as weaker -reflect both societal change and resource constraint. Many precedents, campaigns, and programs encourage female education, and this is augmented by the observation that, in urban Mumbai at least, girls nowadays usually go to school. In this sense, the descriptive norm favours female education -up to a point -and perhaps the injunctive norm is itself changing. Urban life is expensive and people are time-poor. Women need to be able to manage household expenses and income from their own employment. At the same time, the gendered division between reproductive and productive work has historically meant that domestic parsimony and responsibility are characteristics of a good wife and mother. Whether or not there is increasing permissiveness toward premarital sex, the penalties for young women can be severe.
Young men, however, have traditionally been granted the privilege of premarital sexual relationships and, although this licence may have more of a passive than an active character, it is arguably part of their gender role." It also seems like the results from the study in Bihar are not integrated into the rest of the discussion. It does not flow with the paragraph before and after it.

Response
We have cut down the discussion of the Bihar study. Its findings remain relevant and we've tried to position it best within the narrative. The discussion and conclusion provide a great application of the results to the work that is being done by the organization. It would be great if the authors could also add how these results could also be applied to other work being done in India, perhaps citing a few other types of interventions that could also benefit from the findings of this study.

Response
We have added to the end of the discussion: "This idea of expanding the reference group has wide applicability and underlies, for example, participatory programs to improve child survival, nutrition, and adolescent sexual and reproductive health in India and the region, all of which are affected by normative behaviour." We have not competing interests Competing Interests: